This directory contains results updated in real time from an experiment about the partly supervised learning of representations and concepts.

The code is available on https://fleuret.org/git/culture and a summary of the whole experiment is available as a thread on X.com.

plots.png

accuracy and perplexity plots

w_quizzes/

examples of quizzes from the reference hand-designed tasks.

c_quizzes/

quizzes that have been created by the generative models.

prediction/

prediction on quizzes. Green and red frames indicate what has been predicted and if it is correct.

generation/

generation of quizzes.

other/

other files

20240914-19:49:22 wrote ae_002.pth
20240914-19:49:23 wrote ae_003.pth
20240914-19:49:23 wrote ae_004.pth
20240914-19:49:24 wrote ae_005.pth
20240914-19:49:25 wrote ae_006.pth
20240914-19:49:25 wrote ae_007.pth
20240914-19:49:26 wrote ae_008.pth
20240914-19:49:27 wrote ae_009.pth
20240914-19:49:27 epoch_duration 2min49s next_finish 19:52:16
20240914-19:49:27 wrote state.pth
20240914-19:49:27 --- epoch 1514 ----------------------------------------
20240914-19:49:27 current_test_accuracies 0.9672 0.9607 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
20240914-19:49:27 time_train=162 time_c_quizzes=9245
20240914-19:49:27 nb_c_quizzes 3278
20240914-19:49:27 training model 2 (accuracy 0)
20240914-19:49:27 training model 3 (accuracy 0)
20240914-19:51:30 train_loss 1514 model 2 0.010743158443101038
20240914-19:51:31 train_loss 1514 model 3 0.011024087306877846
20240914-19:51:32 test_loss 1514 model 2 0.007481033686613614
20240914-19:51:32 test_loss 1514 model 3 0.006486562787495131